X-Trans ex-noise: Testing DxO’s new denoising tech for Fujifilm cameras: Digital Pictures Evaluate


This picture was captured with a Fujifilm X-T5 at ISO 12,800 and processed utilizing DxO’s DeepPRIME XD denoising algorithm.

It has been a little bit over a decade since Fujifilm launched its first digicam with an X-Trans sensor, the Fujifilm X-Pro1, approach again in January 2012. With a shade filter array that samples crimson, inexperienced and blue pixels in each horizontal line and vertical column, X-Trans sensors are designed to extra evenly pattern the colour data throughout the picture. This, in flip, permits higher resistance to paint moiré versus the extra conventional Bayer filter, which skips both the crimson or blue pixels in each different line or column.

The downsides of the X-Trans method are twofold. With a bigger 6×6 sample moderately than the 2×2 of Bayer, it makes the duty of demosaicing the colour information not simply distinctive to Fujifilm cameras – requiring software program makers to implement extra customized code – but in addition extra processor-intensive and difficult to optimize. And French software program maker DxO‘s equally distinctive method of denoising previous to demosaicing additional difficult the scenario. So for years it merely eschewed X-Trans assist altogether, providing Fujifilm digicam assist just for its Bayer-based fashions.

The denoising samples all through this text have been produced from these 5 photographs, proven right here of their out-of-camera JPEG variations. You may click on every crop within the rollover galleries under to open the full-sized, DxO-processed JPEG.

In October 2021, that lastly began to vary with the introduction of DxO’s PhotoLab 5. That introduced fundamental X-Trans assist, though a few gaps remained by which noise discount algorithms have been out there for it. The current arrival of PhotoLab 6.4 has now closed the largest of those, bringing assist for the corporate’s top-of-the-line DeepPRIME and DeepPRIME XD denoising algorithms to Fujifilm X-series shooters. (This leaves simply the corporate’s commonplace PRIME denoising algorithm and a handful of extra minor options off the desk on your X-series digicam.)

So how does DeepPRIME get on with X-Trans imagery in the true world? Let’s roll up our sleeves and take a more in-depth look, courtesy of a choice of excessive ISO pictures from our current Fujifilm X-series critiques!

All pattern photographs and their crops on this article have been transformed from uncooked utilizing the just-released PhotoLab 6.5 replace (additionally out there for macOS) with its “DxO Optical Corrections solely” preset, and every denoising algorithm was used at its default settings.

Take a look at 1: Bread (Fujifilm X-T4, ISO 2500)

No denoising HQ denoising DeepPRIME DeepPRIME XD

We’ll begin off with a shot from our Fujifilm X-T4 overview at ISO 2500, a reasonably excessive sensitivity. As you’ll be able to see, with no denoising utilized the noise intrudes fairly a bit and, regardless of the X-Trans shade filter array, there’s nonetheless a good bit of false shade artifacting on show too.

The bottom HQ denoising algorithm cleans issues up significantly, however it additionally robs some element from the finer flour particles caught to the bread, and the cracking within the crust appears to be like a bit smooth at 1:1 decision.

And whereas DeepPRIME crispens issues up, the newly-available DeepPRIME XD algorithm positively does a significantly better job. Its result’s even crisper, and it restores extra of the fantastic particulars, yielding a much less plasticky really feel than the usual DeepPRIME.

Take a look at 2: Duck (Fujifilm X-S10, ISO 3200)

No denoising HQ denoising DeepPRIME DeepPRIME XD

Growing the sensitivity a little bit to ISO 3200 with a shot from our Fujifilm X-S10 overview, now we have a tightly-cropped shot of a duck’s head with the feathers freshy wetted from a go to beneath the water. Once more, we will see fairly a little bit of noise previous to denoising, particularly noticeable within the duck’s eye. False shade additionally intrudes, particularly within the reflective water droplets and the fantastic folds of the duck’s decrease eyelid.

Switching to HQ denoising improves each of those defects significantly, however it positively discards a superb little bit of fantastic element too, throwing the child out with the bathwater. You may see this most simply within the actually fantastic feathers beneath and instantly behind the attention.

The DeepPRIME algorithm restores a lot of the element in these areas, however seen 1:1 it additionally provides a considerably unnatural, waxy look. It is noticeable within the feathers to some extent, however particularly within the eyeball, which nearly seems as if fabricated from glass.

Switching to DeepPRIME XD brings again just a bit extra element within the eye and its reflections. It additionally provides extra crispness to the feathers. Once more, this result’s clearly one of the best of the group, particularly taking into consideration it is at default settings with no consumer intervention.

Take a look at 3: Samurai (Fujifilm X-H2S, ISO 5000)

No denoising HQ denoising DeepPRIME DeepPRIME XD

Subsequent up, now we have an ISO 5000 shot of a samurai warrior figurine from our Fujifilm X-H2S overview. With no denoising the shiny, black panels look fairly speckled. And once more, there’s a little bit false shade artifacting, which is particularly noticeable within the shiny space alongside the higher fringe of the small twine clip which you’ll see close to the highest middle of the crop.

The HQ denoising algorithm cleans each of those up fairly properly, however it additionally robs us of nearly all of the thread patterns within the material on the proper of the crop. The threads within the cords are additionally largely misplaced, whereas some unnatural artifacts will be seen within the shiny bronze trim piece on the middle of the crop.

The usual DeepPRIME algorithm does a noticeably higher job with these defects, however it nonetheless appears to be like a little bit smooth when seen 1:1. As soon as once more, DeepPRIME XD appears to be like noticeably crisper, with extra of the threads remaining seen in each material and cords.

Take a look at 4: Canine (Fujifilm X-H2, ISO 6400)

No denoising HQ denoising DeepPRIME DeepPRIME XD

Our subsequent shot is at ISO 6400 and comes from our Fujifilm X-H2 overview. With denoising disabled, it is fairly a snowy affair, particularly with the reflections within the canine’s eyeball. A lot of the element in its fur fairly closely obscured by noise, particularly within the blue channel. And whereas the HQ denoising filter helps, it nonetheless leaves issues fairly desaturated.

With the DeepPRIME filter lively, not solely is that noise principally squashed, however the colours within the fur and the reflections within the canine’s eye are noticeably extra vivid. And that is much more true of the DeepPRIME XD model. The XD algorithm additionally rids the fur of some very fantastic artifacts which might be particularly noticeable within the high proper nook of the usual DeepPRIME model.

Take a look at 5: Avenue (Fujifilm X-T5, ISO 12800)

No denoising HQ denoising DeepPRIME DeepPRIME XD

Up to now, all of those comparisons have performed to the strengths of each DeepPRIME and DeepPRIME XD. As we have seen in our previous protection of those applied sciences – our DxO PhotoLab 6 overview, for instance – DxO’s AI denoising tends to carry out finest when introduced with naturally-occurring element. Issues like feathers, fur and foliage are its strengths.

The place it could possibly typically stumble is with man-made element, and particularly textual content. So for this ISO 12800 shot from our Fujifilm X-T5 overview, we have saved the toughest job for final. A nighttime metropolis road scene, it isn’t solely shot on the highest sensitivity of the bunch but in addition comprises beneficiant portions of each of those AI bugbears.

As we will see above, DeepPRIME XD continues its successful methods with the ornate moldings on the backside proper of the crop, the place your eye is fooled by the crispness and further element, and would not essentially discover any inaccuracies within the rendering. However these methods do not work so nicely with the store signal. (And the identical holds true of the crimson promoting hoarding under, for those who click on via to the full-sized model.)

In each of these areas, the usual DeepPRIME algorithm – whereas noticeably much less sharp than the XD variant – yields extra readable-looking textual content with fewer unpleasant artifacts. As people, we’re programmed to establish issues like faces and writing: we will make these out comparatively simply even when a bit blurred, however we additionally discover defects of their rendering extra readily.

No denoising HQ denoising DeepPRIME DeepPRIME XD

We will have a look at a second crop of this picture, although, to level out some variations elsewhere inside the body. Right here, we’re trying a bit additional to the fitting on the condominium home windows, and there are a few issues to notice.

Firstly, DeepPRIME XD appears to be tuned to search for and try to duplicate patterns extra strongly than the usual DeepPRIME. That distinction is particularly seen within the window blinds; they’re simply barely seen in all however the DeepPRIME XD picture, which renders them with a lot higher distinction.

Secondly, have a look at the roof tiles in direction of the underside of the crop. Within the DeepPRIME XD model, these have a considerably unnatural, virtually painterly look to them that is not current within the DeepPRIME or different variations. This identical impact can also be noticeable within the background of the signal from the primary crop, though there it is a little bit more durable to inform what is going on on.


From my testing, the arrival of DeepPRIME XD for X-Trans appears to be like to be a win for Fujifilm X-series shooters, simply as it’s for these on different platforms. Though it is not all the time going to be your finest wager, it performs higher than the options incessantly sufficient that for those who’ve the time required for its processing, it is price making your default choice for high-ISO pictures.

It is also price noting that even when it struggles with issues like textual content, you normally need to be trying on the picture very carefully to identify the defects. Considered from a extra typical distance, a DeepPRIME XD shot goes to create the impression of extra element and crispness, even when it is to some extent a synthetic invention. When not pixel peeping, that may make the general picture really feel higher even when a extra rigorous examination would possibly inform a special story.

You need to bear your topics, the viewer and the way they will be taking a look at your photographs in thoughts when deciding which algorithm to make use of.

The purpose right here is that it’s a must to bear your topics, the viewer and the way they will be taking a look at your photographs in thoughts when deciding which algorithm to make use of. And we’ll restate that we have solely proven outcomes at default settings; you’ll be able to all the time dial again the power of an algorithm to enhance the outcomes for any given shot.

General, I’ve discovered DeepPRIME XD to be a really useful gizmo and I am thrilled that it is now out there for X-Trans shooters too! In the event you’re not already a PhotoLab buyer, I extremely suggest testing the free 30-day trial with your personal photographs to see if it is price including to your personal digital darkroom.


Please enter your comment!
Please enter your name here